Fine-tuning on theoretical knowledge vs. practical examples

AI Discussions
, ,
1

Hey everyone,

I’ve been diving into the world of fine-tuning language models, especially for creating high-quality courses. Important aspect of high quality courses are: ensuring the course structure is well-organized, the content is engaging, and the complexity of topics gradually increases throughout the course.

During my research I thought about a pretty interesting question: How effective is training models on theoretical knowledge compared to practical examples?

Would a model be able to generate high quality course content by continued pre-training it on theoretical knowledge? This would consist of materials on how to develop and evaluate courses, covering topics like instructional design, learner engagement strategies, and assessment techniques.

Or would it be better to fine-tune it on actual course materials from successful online courses, including lectures, assignments, quizzes, and project examples?

To facilitate discussion, I would like to generalize the questions as follows:

  • Understanding Effectiveness: How does training on theoretical knowledge compare with training on practical examples in terms of the model’s ability to perform a task effectively?
  • Combining Approaches: Can combining theoretical and practical data provide better results? If so, how should this be balanced?

I look forward to hearing your thoughts on this!

2

The model needs to train on both theoretical and

because it also needs to learn the structure of a successful course other than just have theoretical knowledge in it! It could be possible to come up with a nicely arranged course but you also need a lot of training data i.e. successful courses and I am not sure you can find that many but you can try!